Hi, As the current version of Marvin is not getting enough traction in both usage and development, some action needs to be taken. After discussing with a few stakeholders, we believe that the scope of the problem Marvin is trying to solve should be simplified. One of the ideas that I would like to propose is for us to remove some parts of the code, related to training and inference execution in production, and instead generate packages that can be executed by other de-facto platforms out there, argo, kfp, tecton, tfx, etc.
I have no doubts that this would simplify the codebase immensely, and in addition, it would make Marvin useful for more people, by taking advantage of this new integration layer. If enough people are interested in this change, I suggest we schedule a meeting to kick off the ideas for this RFC. Please respond with +1 or -1 Best, Lucas
